Sentiment Domain Adaptation with Multiple Sources

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 301–310, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Sentiment Domain Adaptation with Multiple Sources Fangzhao Wu and Yongfeng Huang∗ Tsinghua National Laboratory for Information Science and Technology Department of Electronic Engineering Tsinghua University, Beijing, China wufangzhao@gmail.com, yfhuang@tsinghua.edu.cn Abstract Domain adaptation is an important re- search topic in sentiment analysis area. Existing domain adaptation methods usu- ally transfer sentiment knowledge from only one source domain to target do- main. In this paper, we propose a new domain adaptation approach which can exploit sentiment knowledge from mul- tiple source domains. We first extrac- t both global and domain-specific senti- ment knowledge from the data of multi- ple source domains using multi-task learn- ing. Then we transfer them to target do- main with the help of words’ sentimen- t polarity relations extracted from the un- labeled target domain data. The similar- ities between target domain and different source domains are also incorporated into the adaptation process. Experimental re- sults on benchmark dataset show the ef- fectiveness of our approach in improving cross-domain sentiment classification per- formance. 1 Introduction Sentiment classification is a hot research topic in natural language processing field, and has many applications in both academic and industrial ar- eas (Pang and Lee, 2008; Liu, 2012; Wu et al., 2015; Wu and Huang, 2016). Sentiment classi- fication is widely known as a domain-dependent task (Blitzer et al., 2007; Glorot et al., 2011). The sentiment classifier trained in one domain may not perform well in another domain. This is because sentiment expressions used in different domain- s are usually different. For example, “boring” ∗Corresponding author. and “lengthy” are frequently used to express neg- ative sentiments in Book domain. However, they rarely appear in Electronics domain (Bollegala et al., 2011). Thus a sentiment classifier trained in Electronics domain cannot accurately predict their sentiments in Book domain. In addition, the same word may convey different sentiments in differen- t domains. For example, in Electronics domain “easy” is usually used in positive reviews, e.g., “this digital camera is easy to use.” However, it is frequently used as a negative word in Movie domain. For instance, “the ending of this movie is easy to guess.” Thus, the sentiment classifier trained in one domain usually cannot be applied to another domain directly (Pang and Lee, 2008). In order to tackle this problem, sentiment do- main adaptation has been widely studied (Liu, 2012). For example, Blitzer et al. (2007) pro- posed to compute the correspondence among fea- tures from different domains using their associa- tions with pivot features based on structural corre- spondence learning (SCL). Pan et al. (2010) pro- posed a spectral feature alignment (SFA) algorith- m to align the domain-specific words from differ- ent domains in order to reduce the gap between source and target domains. However, all of these methods transfer sentiment information from only one source domain. When the source and target domains have significant difference in feature dis- …

Content truncated to 500 words